Nouveau - Oct 2013 - [Bug 70212] New: glxinfo triggers assert in cso_release_all after 3f0627c2ad6

https://bugs.freedesktop.org/show_bug.cgi?id=70212

          Priority: medium
            Bug ID: 70212
          Assignee: nouveau at lists.freedesktop.org
           Summary: glxinfo triggers assert in cso_release_all after
                    3f0627c2ad6
          Severity: normal
    Classification: Unclassified
                OS: All
          Reporter: awatry at gmail.com
          Hardware: All
            Status: NEW
           Version: git
         Component: Drivers/DRI/nouveau
           Product: Mesa

I just switched my GF9400M from Nvidia's blob to nouveau with mesa/drm from
git.

When executing glxinfo with git master, I get the following assert after the
list of core profile extensions are printed:

cso_cache/cso_context.c:315:cso_release_all: Assertion `max <= 16'
failed.
Trace/breakpoint trap (core dumped)


I've bisected the regression to this commit:
3f0627c2ad605b006737312c478907859411ffa8 is the first bad commit
commit 3f0627c2ad605b006737312c478907859411ffa8
Author: Brian Paul <brianp at vmware.com>
Date:   Thu Sep 12 18:09:33 2013 -0600

    nouveau: implement pipe_context::bind_sampler_states()

:040000 040000 b09183c220eb383907e212299a3ed5d66bd26c18
cbfe533aeb6e5ff8648ac8efdf60a28b86ffe5e0 M    src

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20131006/29bd741c/attachment-0001.html>

https://bugs.freedesktop.org/show_bug.cgi?id=70212

--- Comment #1 from Brian Paul <brianp at vmware.com> ---
I can't test nouveau here.  Can you debug this a bit further?  What's
the value
of 'max' and 'sh'?
Looks like the driver is returning an unexpected value for the
PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS query.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20131007/4f51bf9d/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=70212

--- Comment #2 from Emil Velikov <emil.l.velikov at gmail.com> ---
Hi Brian

The nv50 driver returns max=32 for sh=PIPE_SHADER_{VERTEX,FRAGMENT,GEOMETRY}
and max=0 otherwise.

Another interesting point is that with the bind_*_sampler_states hooks the
drivers where left to handle cases of 0 samplers on their own - with nouveau
and radeon doing additional work (state tracking afaics). Whereas with the
unified bind_sampler_states() in case of 0 samplers the driver hook is not
executed. Is that change intentional ?

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20131007/fdb21064/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=70212

Emil Velikov <emil.l.velikov at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED

--- Comment #3 from Emil Velikov <emil.l.velikov at gmail.com> ---
(In reply to comment #2)
> Another interesting point is that with the bind_*_sampler_states hooks the
> drivers where left to handle cases of 0 samplers on their own - with
nouveau
> and radeon doing additional work (state tracking afaics). Whereas with the
> unified bind_sampler_states() in case of 0 samplers the driver hook is not
> executed. Is that change intentional ?
Scratch that

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20131012/f8ebc2cc/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=70212

--- Comment #4 from exc <exclusive_tm at mail.ru> ---
It's because value returned by nv50_screen_get_shader_param with
param=PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS is greater than PIPE_MAX_SAMPLERS.
Setting either of them to another fixes the problem, dunno which is right. 

git blame outputs:

% git blame src/gallium/drivers/nouveau/nv50/nv50_screen.c -L 251,252
f5bfe54a src/gallium/drivers/nv50/nv50_screen.c (Marek Ol??k 2011-09-27
22:22:06 +0200 251)    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
f5bfe54a src/gallium/drivers/nv50/nv50_screen.c (Marek Ol??k 2011-09-27
22:22:06 +0200 252)       return 32;

% git blame src/gallium/include/pipe/p_state.h -L 60,60
da893403 (Brian 2008-04-08 21:43:36 -0600 60) #define PIPE_MAX_SAMPLERS        
16

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20131020/ba2b3bc6/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=70212

--- Comment #5 from Alen Skondro <askondro at gmail.com> ---
I had the same problem with nv50. 
Now that I own an NVE4 the issue is gone, the assert passes.

in src/gallium/drivers/nvc0/nvc0_screen.c

 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
254       return 16; /* would be 32 in linked (OpenGL-style) mode */
255       /*
256    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLER_VIEWS:
257       return 32;
258       */

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20131023/be19b55d/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=70212

--- Comment #6 from Brian Paul <brianp at vmware.com> ---
Created attachment 88040
  --> https://bugs.freedesktop.org/attachment.cgi?id=88040&action=edit
limit samplers returned to PIPE_MAX_SAMPLERS

Can someone verify this patch to nv50?

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20131023/8741b43d/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=70212

--- Comment #7 from Aaron Watry <awatry at gmail.com> ---
(In reply to comment #6)
> Created attachment 88040 [details] [review]
> limit samplers returned to PIPE_MAX_SAMPLERS
> 
> Can someone verify this patch to nv50?
I can confirm that this patch avoids the assert in glxinfo on the original
hardware which encountered the bug (NVAC, GF9400M).

Thanks,
Aaron

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20131023/44401b70/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=70212

Brian Paul <brianp at vmware.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #8 from Brian Paul <brianp at vmware.com> ---
Fixed w/ commit c1345720c86639bb570f3495b834916849d50e37

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20131023/fdec1cb6/attachment.html>